Sicheng Lv | Searcharxiv

arXiv2026

Building evidence-based knowledge bases from full-text literature for disease-specific biomedical reasoning

Biomedical knowledge resources often either preserve evidence as unstructured text or compress it into flat triples that omit study design, provenance, and quantitative support. Here we present EvidenceNet, a disease-specific dataset of record-level evidence collections and corresponding graph representations derived from full-text biomedical literature. EvidenceNet uses a large language model (LLM)-assisted pipeline to extract experimentally grounded findings as structured evidence records, normalize biomedical entities, score evidence quality, and connect related records through typed semantic relations. We release EvidenceNet-HCC with 7,872 evidence records and a corresponding graph with 10,328 nodes and 49,756 edges, and EvidenceNet-CRC with 6,622 records and a corresponding graph with 8,795 nodes and 39,361 edges. Technical validation shows high component fidelity, including 98.3% field-level extraction accuracy, 100.0% high-confidence entity-link accuracy, 87.5% fusion integrity, and 90.0% semantic relation-type accuracy. Downstream analyses show that the data support retrieval-augmented question answering and graph-based tasks such as future link prediction and target prioritization. These results establish EvidenceNet as a disease-specific biomedical knowledge base dataset for evidence-aware analysis and reuse.

cs.CE↗

arXiv2025

RL Is Neither a Panacea Nor a Mirage: Understanding Supervised vs. Reinforcement Learning Fine-Tuning for LLMs

Training large language models (LLMs) from scratch is increasingly impractical, making post-training methods such as supervised fine-tuning (SFT) and reinforcement-learning fine-tuning (RL-FT, e.g., PPO) central to modern practice. Using an out-of-distribution (OOD) variant of the 24-point card game and new spectrum-based diagnostics, we revisit how these two stages reshape model representation and OOD performance. Our key findings are- (1) RL-FT can restore much of the OOD performance loss from SFT (e.g., Llama-11B 8.97% to 15.38%, Qwen-7B 17.09% to 19.66%). But when SFT induces severe overfitting and a clear distribution shift, RL-FT cannot fully recover OOD performance. (2) Direction shifts of singular vectors matter more than singular value magnitudes. These shifts concentrate on directions linked to the largest and smallest singular values, leaving the bulk spectrum intact. (3) Low-rank and shallow recovery is effective: restoring singular vector directions for the top 20% of values or first 25% of layers recovers 70-80% of OOD performance. (4) Stronger SFT checkpoints enable better recovery by RL, while overfitted ones resist restoration. These results reconcile prior reports of RL superior OOD performance: RL primarily counteracts SFT-induced directional drift rather than finding new solutions. Our spectrum-aware analysis highlights inexpensive recovery knobs low-rank UV merging and shallow-layer resets that practitioners can use before costly RL fine-tuning.

cs.LG↗